IoT MALWARE CLASSIFIER

Overview

IoT Smart Home Devices use standard protocols to communicate over the internet. It is possible using tools like Zeek to capture the network traffic of these devices, which we can then analyze using a variety of methods.

The focus of this project centered around using a network capture dataset (IoT-23) to train a robust neural network malware classifier capable of detecting smart home IoT attacks in near real-time.

Results

In summary, the project successfully developed a deep neural network (DNN) based malware classifier for IoT smart home devices that achieved high accuracy (99.897%) on a tested dataset.

Additionally, the project demonstrated the effectiveness of a transfer learning method to update the classifier with new malware data, with minimal impact on performance (0.0638% accuracy reduction when tested on an extended dataset). The project met its objectives and results were extensively documented.

Takeaways

I gained experience with transfer learning and how to update machine learning models with new data - this was a key takeaway for me.

I was able to practice evaluating performance metrics, such as classification accuracy and interpret them. I also gained experience working with datasets, including cleaning, preprocessing and splitting the data for training and testing. In addition, I became familiar with IoT and Network security area, and how to detect the type of malware that target IoT devices.

The project required me to use Python and relevant libraries such as pytorch, fastai, scikit-learn and pandas, which I had a chance to improve on. Lastly, I was reminded the importance of documenting my work, including clear explanations of the objectives, methods, results, and conclusions.